Keywords

1 Introduction

During the last decades, the Web Engineering research community has advocated the use of Model-Driven Web Engineering (MDWE) to improve the development processes for Web applications. MDWE is based on the assumption that using development methods that rely on models with code generation capabilities improves the global developer experience. Among the set of purported MDWE advantages, lower entire product life cycle and maintenance costs, higher end-user satisfaction, shorter time-to-market and less human resources and short and long term productivity gains stand out [1,2,3]. These advantages are increasingly being supported, at least in certain contexts, by empirical evidence [4].

However, in spite of the available data, the paradigm shift from pure code-centric approaches to MDWE that has been expected in industry for years is still to come. We agree with [5] in that this low level of adoption of MDWE approaches may be partly due to the fact that method assessment efforts still mostly revolve around method technological features (such as separation of concerns, the availability of tools or artifacts traceability, to name a few) while paying little attention to the developers’ attitudes and perceptions of the method. This may be caused by the former being much easier to understand and quantify by technical people [6], who may at the same time undervalue the impact of soft skills on their work [5].

Being MDWE development a socio-technical activity, it is these soft skills what determines the final acceptance of the method and, ultimately, of the paradigm. In fact, in Software Engineering (SE), human cognitive limitations and personal behaviour have been known to play an important role in job performance and attitudes since the early seventies [5, 7, 8], to the point that individual performance differences have been pointed out as the largest source of variation in software development performance [9]. For this reason, understanding what distinguishes individual model-driven Web engineers should be an integral part of a well developed theory of MDWE. Moreover, given the magnitude of the individual differences, an understanding of the source(s) of these differences may yield real gains in MDWE programming productivity [10]. Moreover, such study of human factors in MDWE needs to be performed with scientific rigour [11]. In this sense, the provision of theoretical models can contribute to increasing such levels of rigour by structuring knowledge and providing explanations and understanding to observed phenomena in terms of basic concepts and underlying mechanisms [12].

Given this context, the contribution of this paper is twofold: it presents both a theoretical model (TM) that relates personality and attitudes towards the use of MDWE methods, and an observational study that, based on that model, empirically evaluates the impact of personality on the intention to adopt an MDWE method.

The paper is organized as follows: in Sect. 2 we present the main research results regarding both the impact of personality on performance and attitudes and the main models that explain technology adoption behaviour. Section 3 builds up on the related work and presents a TM that, based on previous proposals, summarizes the variables that, according to the existing knowledge, may impact user attitudes regarding MDWE adoption. Based on such TM, Sect. 4 presents the planning, execution and data analysis of an observational study in which we have analyzed the relationship between a set of personality traits and the intention to adopt an MDWE method. Last, Sect. 5 presents the main conclusions and some further lines of research.

2 Related Work

As it was previously mentioned, behavioural (i.e. human and organizational) factors have shown to have a substantial impact on job performance and attitudes [13]. This suggests that software companies are particularly vulnerable to people problems [14]. Despite this recognition, however, the SE community has traditionally neglected people factors [15], which in turn has had a detrimental impact on progress in SE [16].

There are two main groups of individual characteristics that have demonstrated an impact over job preferences and performance: personality and cognitive abilities [17]. Previous studies have repeatedly found that there is very little correlation between personality and cognitive abilities, which supports the idea of them being two different, unrelated, constructs [18].

In this paper we center on personality, which is considered to be a relatively stable trait [19].

2.1 Main Personality Theories and Their Impact on SE Activities

According to the APA [20], personality can be defined as individual differences in characteristic patterns of thinking, feeling, and behaving. Personality matters because it predicts and explains behavior and performance at work [21]. People with different personalities will regard a problem from different perspectives and make different decisions in a MDWE project [15].

A mapping study from 2015 [11] detected how the personality concept can be worked on according to seven different perspectives, among which the dispositional one, which encompasses the traits and types theory, is the one preferred in organizational psychology and in studies on personality in SE. Such trait and type approach assumes that personality consists of stable inner qualities that differ between individuals and influence behavior.

There are three mainstream theories competing for the attention of SE researchers: (1) the Jung’s Personality Types theory [22], (2) the Big Five (BF) personality model [23], and (3) the Eysenck’s hierarchical three factor model [24], also referred to as the Giant Three Model or the PEN model [25]. From them, BF and PEN are considered to be the two theories better representing the personality structure [26]. The BF model includes five factors: extroversion, openness to experience, agreeableness, conscientiousness and neuroticism. The PEN model limits these factors to three: extraversion/introversion, neuroticism/stability and psychoticism/socialization. Also, they have different aims: while the aim of BF is the classification of all major sources of individual differences in personality, the PEN model has the primary aim of identifying those dimension of individual differences which are well grounded in biological processes [27]. Several studies have compared the suitability of the PEN model vs the BF model [27]. They conclude that Eysenck’s E and N correspond to BF’s E and N dimensions. Eysenck’s P dimension has shown to overlap with A and C from the Big Five. Last, BF’s O, which is the fifth dimension in BF, is also considered to be a component of P in Eysenck’s PEN model. Both models have shown similar psychometrical validity, and the discussion remains open regarding whether three or five factors best represent the basic dimensions of personality [25]. For the sake of parsimony, in this paper we have chosen the PEN model as the basis for our theoretical framework.

As far as SE activities are concerned, the research community has studied the impact of personality on very different domains [11]. From them, the relationship between personality skills and roles in a software team (personality-job fit) stands out. In this line of research, several studies outline the personality traits that, depending on the role the subject plays in the development team, goes the furthest towards explaining successful outcomes [28]. Closer to our topic of interest in this paper, some researchers have studied the relationship between personality traits and job preferences (regardless of the success in performing the set of tasks involved in such job) [28]. The results point at extrovert people preferring analyst jobs, while introverts opting for designer or programming roles. However, there is a research gap regarding the impact of personality on intention to adopt SE methods and techniques. This gap extends to the relationship between personality and intention to adopt MDWE approaches.

2.2 Intention to Adopt SE Methods

During the last years plenty of theoretical models have emerged aiming at clarifying the factors affecting the adoption of information technologies, tools and methods by individuals and organizations [29]. Most of these models adapt, either directly or indirectly, the Technology Acceptance Model (TAM) [30]. Particularly, several studies have applied variations of the TAM to evaluate Model-Driven Engineering (MDE) approaches. Walderhaug et al. [31] have used TAM to measure the intention to adopt MDE in the healthcare domain. Their findings suggest that perceived usefulness and ease of use are the most important factors for adopting MDE. No significant relationship between future use intentions and tool performance or subjective norm were found. Condori et al. [32] have also applied a modified version of TAM to analyze the acceptance of a model-based measurement procedure. Their results show that the intention for using their measurement method is more strongly influenced by the perceived usefulness than by the perceived ease of use. Finally, Mohagheghi et al. [33] apply an extended version of the TAM model for the evaluation of model-driven approaches in an industrial context. They conclude that usefulness, ease of use and perceived tool maturity are important determinants for the adoption of MDE in industry.

In an effort to come up with a model that summarizes all this knowledge and overcomes some of their well-known problems (lack of proper definitions for the dimensions, use of the same names for different purposes, confusion among types of variables, etc.), in a previous work we proposed a Unified Method Adoption Model (UMAM) [29]. It decomposes the intention to adopt construct into the five dimensions on which the research community has reached a wider consensus: usefulness, ease of use, subjective norm, compatibility and voluntariness.

Next, we present a theoretical framework that puts together all these constructs.

3 Theoretical Model

In setting an agenda for the empirical assessment of the impact of human factors on MDWE adoption intention, the basis should be a theoretical framework; its existence greatly contributes to the soundness of the empirical work, since it provides guidance in targeting what behaviour to study and in assessing a study’s results [10, 34]. Unfortunately, to the best of our knowledge, such theoretical model does not exist so far, so the first contribution of this paper is the proposal of a framework (see Fig. 1) that, based on the state of the art in the field, defines personality factors, intention to adopt factors, and their main relationships.

Fig. 1.
figure 1

Theoretical framework: personality factors influencing MDE adoption intention.

Full size image

As it was explained in Sect. 2, for the definition of personality we have abided by the PEN model, while, for the intention to adopt definition, we have used the UMAM model. In Fig. 1 we can observe the different constructs for (a) personality and (b) intention to use, whose definitions are as follows:

  • Personality. It is made up of four dimensions: Extraversion/introversion (E), Neuroticism/Stability (N), Psychoticism/Socialisation (P) and Lie/Social Desirability (L).

    • E: Degree to which a person is an extrovert. Higher scores in this dimension imply being more outgoing, talkative, and more in need of external stimulation.

    • N: Degree to which a person shows neurotic attitudes. Higher scores in this dimension imply being more prone to depression and anxiety, and they are more obsessive. Neurotic people have lower tolerance to stress and uncertainty, and the are easily nervous or upset.

    • P: Degree to which a person shows psychotic attitudes. Higher scores in this dimension imply higher levels of assertiveness, aggression, impulsiveness, orientation towards achievement, or tough-mindedness. This dimension is also related to higher levels of creativity and divergent thinking.

    • L: Degree to which a person is more/less likely to show a social/desirability bias. Higher scores in this scale imply higher levels of honesty of the respondent.

  • Intention to adopt. It is made up of five dimensions: Utility (U), Ease of Use (EoU), Compatibility (C), Social Norm (SN) and Voluntariness (V). All these factors load into a global Intention to Use (IU) factor.

    • U: Degree to which a person believes that using a particular method will enhance his/her job performance.

    • EoU: degree to which a person believes that using a particular method or tool would be free of effort.

    • C: Degree to which a method is perceived as being consistent with existing values, principles, practices and the past experience of potential adopters.

    • SN: Degree to which developers think that others who are important to them think they should use a method.

    • V: Extent to which potential adopters perceive the adoption decision to be non-mandated, and represents the felt pressure from the social environment on the potential adopter.

    • IU: Global intention to adopt the method in the future if given the chance.

Our model also determines that the three main personality dimensions (E, N and P) are related with the developer’s perception regarding U, EoU and C of MDWE methods. For example, MDWE methods that allow for lower code control are thought to augment anxiety, so they may be less likely to be adopted (lower perceived U, EoU and C) for subjects scoring higher in N. Similarly, more extroverted people may regard MDWE methods more optimistically, and may also score higher on U, EoU and C. Last, since MDWE methods are oriented towards achievement, people scoring higher on P may regard the methods as more useful and compatible. All this in turn affects the subjects’ global IoA for the method. To the best of our knowledge there is no empirical evidence to suggest that personality affects the Subjective Norm or the Voluntariness of the method, which is why in our model they have no relationship with the perception variables (see Fig. 1).

4 Description of the Observational Study

Observational studies are a kind of empirical study where, contrary to experiments or quasy experiments, the independent variables are not manipulated but rather observed, and, based on such observations, the researcher tries to draw some conclusions [35]. They provide a way to scientifically observe phenomena and report data in situations such as ours in which we cannot manipulate the variables (neither the personality of the developers nor their feelings with respect to MDWE methods). As their main disadvantage, they do not permit to establish cause-effect relationships, but only assess correlations. Next we describe our experimental design and main outcomes.

4.1 Goals and Context Definition

Following the GQM template [36], our empirical study is aimed at analyzing the intention to adopt an MDWE method for the purpose of evaluating its relationship with respect to the personality traits from the point of view of junior software developers. Our research question can be formulated as follows: Are personality traits related with the intention to adopt an MDWE method?

4.1.1 Subjects

The subjects were 71 students enrolled in the ‘Software Design’ course (5th semester of the Computer Science degree) at the University of Alicante. During that course, the students developed a software project following the OOH4RIA method [37]. At the end of the term, they were asked to fill in two questionnaires. From the 71 subjects, 16 subjects did not complete the questionnaires in full, and therefore have been discarded from the analyses. The final sample contained 55 subjects, 10 women and 45 men, whose mean age was 21.6 years old.

4.1.2 MDWE Method

In order to measure the intention to adopt a MDWE method, the students were trained in the OOH4RIA method for 15 h. The OOH4RIA approach extends the OOH method [38] and proposes a complete development process based on a set of models and transformations that allow to go from conceptual models to code. OOH4RIA is also equipped with an Implementation Development Environment (IDE) [39] that offers support for both the design activities and the automatic code generation process. OOH4RIA is free of charge for universities, and it is taught as part of the Computer Science degree in which the empirical study took place.

4.2 Study Planning

The context in which our study took place limited the set of variables to measure. Namely, voluntariness did not apply to our course environment, where all subjects were required to use OOH4RIA.

4.2.1 Questionnaires

In order to operationalize the constructs included in our TM, we selected the following two questionnaires:

  • EPQ-R: Spanish version [40]. This questionnaire includes 83 items divided into four scales: Extraversion/introversion (E, 19 items), Neuroticism/Stability (N, 23 items), Psychoticism/Socialisation (P, 23 items) and Lie/Social Desirability (L, 18 items).

  • UMAM-Q [29]: This questionnaire, which shares many features with questionnaires such as [41, 42], was developed to specifically measure SE method adoption. It includes 42 items divided into six scales: Utility (U, 7 items), Ease of Use (EoU, 7 items), Compatibility (C, 7 items), Social Norm (SN, 7 items), Voluntariness (V, 7 items) and global Intention to Use (IU).

4.2.2 Hypotheses Description

Based on the TM of Fig. 1, we have established the following hypotheses:

  • H1: E is significantly related with the subjects’ perception of U, EoU, C and global IoA

  • H2: E is significantly related with the subjects’ perception of SN

  • H3: N is significantly related with the subjects’ perception of U, EoU, C and global IoA

  • H4: N is not significantly related with the subjects’ perception of SN

  • H5: P is significantly related with the subjects’ perception of U, EoU, C and global IoA

  • H6: P is not significantly related with the subjects’ perception of SN

4.3 Data Analysis

4.3.1 Questionnaire Validation

Table 1 presents the Cronbach’s \(\alpha \) coefficient for both the EPQ-R and the UMAM-Q scales. In EPQ-R, E and N show a high reliability. These results are aligned with the reported reliability of the published questionnaire [40], where the reliability of E is between 0.80 (women) and 0.82 (men), and the reliability of N is 0.86 for both men and women. However, P shows a low degree of reliability. In the questionnaire report the reliability of P is also rather low (around 0.76), but not as low as in our results. As for the UMAM-Q, all the scales show a Cronbach’s \(\alpha \) higher than 0.7, which is proof of sufficient internal consistency.

Table 1. Reliability of EPQ-R and UMAM-Q Scales
Full size table

4.3.2 Descriptive Statistics

In Table 2 we present the descriptive statistics (Mean and Standard Deviation) of all the scales.

Table 2. Descriptive Statistics for the scales
Full size table

4.3.3 Hypothesis Analysis

In order to analyze the relationship between the different personality dimensions and the UMAM variables, for each personality dimension we have classified the subjects into three groups: high (roughly the 27% top scores for each scale), medium (46% middle values) and low (27% lowest scores for each scale). Then we have rounded up or down to the closest score at each cut point, and we have discarded those subjects belonging to the ‘medium’ level. Tables 3, 4 and 5 present the analysis of the data obtained from the questionnaires for the E, N and P variables, respectively. Then, an independent-samples t-test was conducted to compare intention to adopt OOH4RIA in high and low conditions for all the personality variables. Homogeneity of variance was checked with Levene’s test, which showed p values greater than 0.05 for all the analyses.

Table 3. Impact of E on Intention to Adopt the OOH4RIA method
Full size table

For the E scale, we have classified as extroverts those that have a score of 16 or more on the scale (15 subjects out of 55). Introverts, on the other hand, are those who have scored 8 or less (16 subjects out of 55).

If we look at Table 3, we can see how extroverts regard the OOH4RIA method as more useful and more easy to use, which partly aligns with H1. However, such differences are not significant (see results of the t-tests in Table 3). Also, the perceived C seems not to be related with the level of extroversion. Regarding H2, SN seems not to be related with E, which is aligned with the idea underlying the hypothesis.

Table 4. Impact of N on Intention to Adopt the OOH4RIA method
Full size table

For the N scale, we have classified as neurotics those that have a score of 16 or more on the scale (15 subjects out of 55). Stable subjects, on the other hand, are those who have scored 7 or less (18 subjects out of 55).

In Table 4, we can see how more stable developers regard the OOH4RIA method as more useful and more easy to use, and also slightly more compatible with their way of working. This is aligned with the idea behind H3. If we now look at the t-test results in Table 4, we can observe how the U and C differences are not significant, but the EoU is. This means that there is a significant relationship between higher levels of stability and higher perceived EoU of the OOH4RIA method. Regarding H4, again SN seems not to be related with N, which is aligned with the idea underlying the hypothesis.

Table 5. Impact of P on Intention to Adopt the OOH4RIA method
Full size table

Last, for the P scale, we have classified as psychotic those that have a score of 8 or more on the scale (13 subjects out of 55). Social subjects, on the other hand, are those who have scored 4 or less (15 subjects out of 55).

In Table 5, we can see how more social people regard the OOH4RIA method as slightly more useful and easier to use, but slightly less compatible with their way of working. This is aligned with the idea behind H5. If we regard the t-test results, we can observe how none of these differences are significant. Regarding H6, again SN seems not to be related to P, which is aligned with the idea underlying the hypothesis.

4.4 Threats to Validity of the Study

We have followed the classification proposed by Cook and Campbell [43] for the classification of the main threats to the validity of the study: internal, external, construct and conclusion. Threats to internal validity are concerned with the possibility of hidden factors which provide alternative explanations for the result. The main threat to the internal validity of this study is the fact that 16 subjects out of 71 chose not to answer the questionnaires. Although, when asked about why they hadn’t answered, they put forward lack of time reasons, we cannot be sure that their personality profile did not have any impact on such decision. This risk is unfortunately unavoidable, since ethical reasons required the fulfillment of the questionnaires to be voluntary, and the load of work at the end of the term tends to be high. Another internal threat is the low internal validity of the P scale. Threats to external validity are concerned with generalization of the results. Our main threat to the external validity is that subjects are students of the Computer Science degree, which is unrepresentative of the population of MDWE developers in the industry. Also, the use of a specific MDWE methodology like OOH4RIA and its tool constitutes a limited environment and can influence the results. Threats to construct validity refer to the relationship between theory and observation. Both the theory and the scales used to measure the different constructs are widely accepted by the research community, so we do not appreciate any important construct validity. Last, threats to conclusion validity refer to the relationship between the treatment and the outcome. All the statistical analyses have been preceded by tests that assured that the assumptions of the statistical procedure were not being violated.

5 Conclusions and Future Lines of Research

This paper has presented both a TM for the study of the impact of human factors on MDWE adoption intention and an observational study to assess the correlations between personality and intention to use dimensions. According to our TM, extroverted developers regard MDWE methods more optimistically, which is correlated with higher scores on U, EoU and C. Our data (see Table 3) seems to confirm higher ratings for U and EoU, but not for C. Also, none of the results are significant.

Our TM also suggest that stable (as opposed to neurotic) people shows more positive attitudes towards the adoption of MDWE methods. Again, our data confirms such relationship for U and EoU, but not for C. Of particular interest is the statistically significant differences in EoU between developers scoring high and low in N. One possible explanation is the learning curve of MDWE methods and the perceived lack of control of the produced code [44]. Neurotic people (characterized by presenting a lower tolerance to uncertainly) may more acutely suffer these MDWE disadvantages. It is important to note how such impact is quite big, to the point of making differences in IoA nearly statistically significant (see Table 5).

Last, our TM suggests that, since MDWE methods are oriented towards achievement, people scoring higher on P may regard the methods as more useful, easy to use and compatible. However, our data (see Table 5) does not support these claims, although the low internal reliability of the scale prevents us from drawing any conclusions out of these results.

We are conscious that our proposed TM is only a starting point, and that the research community needs to gather much more empirical evidence in order to evolve it to reflect the reality of the MDWE method adoption. In this sense our paper aims at fostering the discussion among the research community regarding individual differences and how they may be affecting the impact of MDWE methods and tools.

As future work we plan to carry out replications of this empirical study, if possible with different MDWE approaches and audiences, in order to augment the external validity of the results and provide more empirical data to validate and, if necessary, evolve the proposed TM.